Global context recovery

Abstract

Methods for error recovery in parsing a deterministic, context-free language need not be derived by ad hoc means. When fixed, table-driven parsing algorithms are used the error recovery procedures can also make use of knowledge about the grammar for the language, derived from the parsing tables. However, at the point of detection of certain kinds of syntax errors, the knowledge of a sentential prefix already parsed, as reflected in the current state of the parsing stack, will be incorrect. Effective error recovery under such circumstances has proven to be difficult. We suggest the use of fiducial, or trustworthy symbols of a language as a key to recovery in these difficult cases. In this paper we give a characterization of such symbols in terms of properties of a grammar and give an algorithm that is capable of global context recovery under all circumstances. It has been implemented in conjunction with an LL(1) parser. We have compared it experimentally with the LR-parser recovery algorithm of Pennello and DeRemer on a test set of 126 small Pascal programs.